Suitable for:

Description:

In this project you can explore approaches to automatically generating iconic gestures for computer-animated characters. Iconic gestures can represent atomic concepts like direction (up, down, left, right) or combined concepts “the small ball” (size, shape). Automatic generation of iconic gestures is highly challenging due to large inter-individual variation and the degree of abstractness. This overall topic can be confined to following tasks:

1. Learning a variety of gestures for atomic concepts and applying the model to new domains, combining the atomic concepts to generate more complex representations (learning weakly compositional structures).
2. Atomic concepts (like shape, size etc.) can be induced from gestures representing combined meanings. Despite large inter-individual variations, there should be some salient features (like direction of movement in a gesture containing “up”) present in every gesture. The goal is to automatically find these features in a corpus and evaluate them in forming new gesture combinations.

Requirements:

• interest in gesture generation
• experience in programming and machine learning
• (optional) experience in machine learning

Contact: